diff --git a/convert.py b/convert.py
index f57363d..29a3aa3 100644
--- a/convert.py
+++ b/convert.py
@@ -35,29 +35,34 @@ def main():
         model.to(device)
     print('done')
 
-    train_filelist, val_filelist = dataset.train_val_split(
+    cache_dir = 'sr{}_hl{}_nf{}'.format(args.sr, args.hop_length, args.n_fft)
+    filelist = dataset.raw_data_split(
         dataset_dir=args.dataset,
-        split_mode=args.split_mode,
-        val_rate=0,
-        vocals=False
+        split_mode=args.split_mode
     )
 
-    filelist = train_filelist + val_filelist
-    cache_dir = 'sr{}_hl{}_nf{}'.format(args.sr, args.hop_length, args.n_fft)
-
     for mix_path, inst_path in filelist:
         X_basename = os.path.splitext(os.path.basename(mix_path))[0]
         y_basename = os.path.splitext(os.path.basename(inst_path))[0]
+        pv_basename = X_basename + '_PseudoVocals'
+        pi_basename = X_basename + '_PseudoInstruments'
+
         print('converting {}...'.format(X_basename))
 
-        X_cache_dir = os.path.join(os.path.dirname(mix_path), cache_dir)
-        y_cache_dir = os.path.join(os.path.dirname(inst_path), cache_dir)
-        v_dir = os.path.join(os.path.split(os.path.dirname(inst_path))[0], 'vocals')
-        v_cache_dir = os.path.join(v_dir, cache_dir)
+        X_dir = os.path.dirname(mix_path)
+        y_dir = os.path.dirname(inst_path)
+        pv_dir = os.path.join(os.path.split(y_dir)[0], 'pseudo_vocals')
+        pi_dir = os.path.join(os.path.split(y_dir)[0], 'pseudo_instruments')
+
+        X_cache_dir = os.path.join(X_dir, cache_dir)
+        y_cache_dir = os.path.join(y_dir, cache_dir)
+        pv_cache_dir = os.path.join(pv_dir, cache_dir)
+        pi_cache_dir = os.path.join(pi_dir, cache_dir)
 
         os.makedirs(X_cache_dir, exist_ok=True)
         os.makedirs(y_cache_dir, exist_ok=True)
-        os.makedirs(v_cache_dir, exist_ok=True)
+        os.makedirs(pv_cache_dir, exist_ok=True)
+        os.makedirs(pi_cache_dir, exist_ok=True)
 
         X, sr = librosa.load(
             mix_path, sr=args.sr, mono=False, dtype=np.float32, res_type='kaiser_fast')
@@ -73,19 +78,20 @@ def main():
         y = spec_utils.wave_to_spectrogram(y, args.hop_length, args.n_fft)
 
         sp = inference.Separator(model, device, args.batchsize, args.cropsize)
-        _, v_spec = sp.separate_tta(X - y)
-        # a_spec, _ = sp.separate_tta(X - y)
+        _, pv = sp.separate_tta(X - y)
+        # pa, pv = sp.separate_tta(X - y)
+
+        # pi = y + pa
 
-        # print('inverse stft of pseudo instruments...', end=' ')
-        # pseudo_inst = y + a_spec
-        # print('done')
+        # wave = spec_utils.spectrogram_to_wave(pv, hop_length=args.hop_length)
+        sf.write('{}/{}.wav'.format(pv_dir, pv_basename), [0], sr)
+        # wave = spec_utils.spectrogram_to_wave(pi, hop_length=args.hop_length)
+        sf.write('{}/{}.wav'.format(pi_dir, pi_basename), [0], sr)
 
         np.save('{}/{}.npy'.format(X_cache_dir, X_basename), X.transpose(2, 0, 1))
         np.save('{}/{}.npy'.format(y_cache_dir, y_basename), y.transpose(2, 0, 1))
-
-        wave = spec_utils.spectrogram_to_wave(v_spec, hop_length=args.hop_length)
-        sf.write('{}/{}_Vocals.wav'.format(v_dir, X_basename), wave.T, sr)
-        np.save('{}/{}_Vocals.npy'.format(v_cache_dir, X_basename), v_spec.transpose(2, 0, 1))
+        np.save('{}/{}.npy'.format(pv_cache_dir, pv_basename), pv.transpose(2, 0, 1))
+        # np.save('{}/{}.npy'.format(pi_cache_dir, pi_basename), pi.transpose(2, 0, 1))
 
 
 if __name__ == '__main__':
diff --git a/lib/dataset.py b/lib/dataset.py
index 2ba7c40..c6120ab 100644
--- a/lib/dataset.py
+++ b/lib/dataset.py
@@ -176,12 +176,12 @@ def make_pair(X_dir, y_dir, v_dir=None):
     return filelist
 
 
-def train_val_split(dataset_dir, split_mode, val_rate, val_filelist=[], vocals=True):
+def train_val_split(dataset_dir, split_mode, val_rate, val_filelist=[]):
     if split_mode == 'random':
         filelist = make_pair(
             os.path.join(dataset_dir, 'mixtures'),
             os.path.join(dataset_dir, 'instruments'),
-            os.path.join(dataset_dir, 'vocals') if vocals else None
+            os.path.join(dataset_dir, 'pseudo_vocals')
         )
 
         random.shuffle(filelist)
@@ -198,22 +198,42 @@ def train_val_split(dataset_dir, split_mode, val_rate, val_filelist=[], vocals=T
     elif split_mode == 'subdirs':
         if len(val_filelist) != 0:
             raise ValueError('`val_filelist` option is not available with `subdirs` mode')
-        
+
         train_filelist = make_pair(
             os.path.join(dataset_dir, 'training/mixtures'),
             os.path.join(dataset_dir, 'training/instruments'),
-            os.path.join(dataset_dir, 'training/vocals') if vocals else None
+            os.path.join(dataset_dir, 'training/pseudo_vocals')
         )
 
         val_filelist = make_pair(
             os.path.join(dataset_dir, 'validation/mixtures'),
             os.path.join(dataset_dir, 'validation/instruments'),
-            os.path.join(dataset_dir, 'validation/vocals') if vocals else None
+            os.path.join(dataset_dir, 'validation/pseudo_vocals')
         )
 
     return train_filelist, val_filelist
 
 
+def raw_data_split(dataset_dir, split_mode):
+    if split_mode == 'random':
+        filelist = make_pair(
+            os.path.join(dataset_dir, 'mixtures'),
+            os.path.join(dataset_dir, 'instruments'),
+        )
+    elif split_mode == 'subdirs':
+        train_filelist = make_pair(
+            os.path.join(dataset_dir, 'training/mixtures'),
+            os.path.join(dataset_dir, 'training/instruments'),
+        )
+        val_filelist = make_pair(
+            os.path.join(dataset_dir, 'validation/mixtures'),
+            os.path.join(dataset_dir, 'validation/instruments'),
+        )
+        filelist = train_filelist + val_filelist
+
+    return filelist
+
+
 def make_padding(width, cropsize, offset):
     left = offset
     roi_size = cropsize - offset * 2